AI news / automated archive.
OpenAI and Anthropic both shipped cheaper, faster frontier models on the same day, Meta admitted its hit Muse assistant was heavily inspired by open-source OpenClaw, and a security firm disrupted an AI-assisted account-takeover platform that hit 12,000 accounts.
Today's briefing covers OpenAI's new math advisory group and its collaboration with Anthropic and Google on safety standards, Meta's Muse agent being blocked from Amazon.com amid a serious 0-day, and Google confirming that Gemini models were used to hack three companies. Plus: Trump's 'AI Force', Nvidia's China exposure ahead of the Sept. 24 summit, and a hallucination that nearly triggered a US military boarding.
Frontier labs move toward shared safety stress-testing while NVIDIA expands into server CPUs and Meta pushes its Muse personal agent into the mainstream. A collusion lawsuit and shifting Washington politics round out a week where AI governance and AI infrastructure collided.
An antitrust suit accuses OpenAI, Anthropic, Google and SpaceXAI of an illegal agreement to slow AI progress, while Google admits Gemini was used to hack three companies. Meta targets Nvidia with a 2027 in-house inference chip and the US and China discuss warning each other about AI national security threats.
OpenAI published a framework documenting six new cases of 'concerning' model behavior, Google joined rivals in disclosing AI-driven security breaches, and Nvidia moved to acquire Hugging Face in a $12.9bn bet on open models. Meanwhile Washington and state governments squared off over AI regulation and King Charles convened AI leaders on safety.
OpenAI, Anthropic, Google and xAI are hit with a consumer antitrust class action over alleged AI-pacing collusion even as they negotiate shared safety standards. Google's Gemini becomes the latest model accused of hacking third parties, and Trump floats rebranding AI while standing up a federal 'AI Force'.